Customize a Shared Bot

You installed a shared bot. Now make it yours in five edits — without breaking what the author got right.

A shared bot is a good default wearing someone else’s life. The author’s schedule, tone, and priorities are baked in. Your job after installing is a five-edit fitting session — keeping the structure, swapping the assumptions.

Edit 1: Schedule and timezone

The author checks at 7am Eastern; you don’t. Reset every time trigger to your day, and state the timezone explicitly — routines anchored to “my morning” drift across seasons otherwise.

Edit 2: Report destination and format

Most shared bots report to chat because chat is universal. If you’d rather read it elsewhere, say where — and cap the format while you’re there:

Report at most N items, M bullets each. If nothing new, one line: “All quiet.”

The quiet-day rule is the edit with the best attention-per-word ratio in bot-building.

Edit 3: Language and audience

Add one line: “Write to me in Chinese; keep source quotes in their original language.” The bot’s instructions can stay in English — models follow English instructions reliably — while everything it produces lands in your language.

Edit 4: The scope trim

Shared bots accumulate breadth to please everyone. Cut to your use:

You only handle「your slice」. Ignore everything else in config.md; don’t report on it.

Every removed responsibility is attention you get back. If the bot came with sources you’ll never read, delete them from the config — file rot starts with someone else’s list.

Edit 5: The boundary check

Confirm the escalation boundary matches your risk appetite, not the author’s:

Restate your read/write permissions and what you will never do without asking.

Compare the answer to what the bot actually does in week one. Mismatch = edit, or uninstall. This overlaps with security review — do both once, it takes ten minutes.

What to keep untouched

• The role definition. The author iterated it; respect the core.
• The state-file layout. Renaming files mid-life breaks routines that reference them.
• The reporting cadence logic — the “how,” not the “when.” Authors usually found the least-noisy shape through real use.

Keep a changelog

One line per edit, dated, in the bot’s workspace:

changelog.md
2026-09-15 — report to Chinese, cap at 5 items
2026-09-20 — removed 3 sources I never read

When the author ships an update, you diff against your changelog and cherry-pick. When the bot misbehaves, the changelog is the first suspect list. And when your edits stabilize, you’re one screenshot away from being the next sharer in the bots directory — the cycle that built the whole ecosystem.